Add checkpoint recovery to scoring and subindexes. This will enable efficient job restarts when something fails.